张新斌 康昌春 李文珲

摘 要：在最近几年中，医学影像技术发展迅猛，进入到大数据的新阶段。怎样从大量的医学图像数据内筛选出重要的信息，显然是医学辨识图像过程中的一个挑战。深度学习属于机器学习兴起的新领域。经由人脑的模拟构建起分层模型，它可以有效地化解传统式机器学习法无法挑选出医学图像内所包藏的信息内容，体现出不可小觑的智能化特征提取、建构复杂化的模型结构以及有效的特征表达性能。更为关键的是，深度学习法可以通过像素级的最初数据逐步地通过底层至高层的途径来提取特征，其为化解辨识医学图像所碰到的新问题指出了新的方向。本论文阐述深度学习的概念，简述主要的模型结构，以乳腺肿瘤X线图像数据的归类为例，研究基于深度学习网络探析医学影像大数据的相关课题。

关键词：深度学习；医学影像；大数据

中图分类号：R445；TP18 文献标识码：A 文章编号：2096-4706（2018）08-0084-03

Abstract：In recent years，medical imaging technology has developed rapidly and has entered a new stage of big data. How to filter important information from a large number of medical image data is clearly a challenge in the process of medical identification. Deep learning belongs to the new field of machine learning. Based on the simulation of human brain，a hierarchical model is constructed. It can effectively resolve the traditional machine learning method，which can not pick out the information contained in medical images，and embody the intelligent feature extraction，construction complex model structure and effective feature expression performance，and more critical is the depth study. The method can extract features through the initial data from the first to the high level through the initial data of the pixel level，which is a new direction for resolving the new problems encountered in the identification of medical images. This paper expounds the concept of deep learning and describes the main model structure. It takes the classification of the X-ray image data of breast tumor as an example，and analyses the related subjects of the big data of medical images based on the depth learning network.

Keywords：deep learning；medical image；big data

0 引 言

从2006年迄今，深度学习即被当作机器学习范围的重要分支诞生。它应用数层的复杂结构或通过数重非线性的变换组成数个处理层，并对数据加以处理。在最近几年中，深度学习广泛地在语音与音频辨识、计算机视觉、自然语言处理以及生物信息学等范围均获得了明显的成果。由于深度学习广泛地运用于数据的分析方面，具有极为可观的运用前景，已被赞誉成2013年迄今的十大最为重要的一项突破性技术。

医疗关系着百姓的生命健康。……